From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:58440) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1fSoGU-0005gy-7f for qemu-devel@nongnu.org; Tue, 12 Jun 2018 14:43:35 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1fSoGQ-0003Ya-AN for qemu-devel@nongnu.org; Tue, 12 Jun 2018 14:43:34 -0400 Received: from goliath.siemens.de ([192.35.17.28]:51223) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1fSoGP-0003XB-Ub for qemu-devel@nongnu.org; Tue, 12 Jun 2018 14:43:30 -0400 References: <267f42a5-b7ce-379e-ffd1-f2611393d2ff@web.de> <6730e270-9906-a43c-68b0-7a09a0743fa5@siemens.com> <7bb24572-0fc1-00de-3552-aca1111059ed@amsat.org> <63ee188e-0c66-ce7e-03b2-fd58e4da9116@siemens.com> <3a128f8d-46a3-591f-ee09-8c9a3f2401ea@amsat.org> From: Jan Kiszka Message-ID: Date: Tue, 12 Jun 2018 20:43:22 +0200 MIME-Version: 1.0 In-Reply-To: <3a128f8d-46a3-591f-ee09-8c9a3f2401ea@amsat.org> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH] e1000e: Do not auto-clear ICR bits which aren't set in EIAC List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: =?UTF-8?Q?Philippe_Mathieu-Daud=c3=a9?= , Dmitry Fleytman Cc: Peter Maydell , qemu-devel , Alexander Graf On 2018-06-12 20:38, Philippe Mathieu-Daud=C3=A9 wrote: > On 06/12/2018 03:30 PM, Jan Kiszka wrote: >> On 2018-06-12 20:11, Philippe Mathieu-Daud=C3=A9 wrote: >>> Hi Jan, >>> >>> On 06/12/2018 02:22 PM, Jan Kiszka wrote: >>>> On 2018-05-22 09:00, Jan Kiszka wrote: >>>>> On 2018-04-16 17:29, Peter Maydell wrote: >>>>>> On 16 April 2018 at 16:25, Jan Kiszka wro= te: >>>>>>> On 2018-04-01 23:17, Jan Kiszka wrote: >>>>>>>> From: Jan Kiszka >>>>>>>> >>>>>>>> The spec does not justify clearing of any E1000_ICR_OTHER_CAUSES= when >>>>>>>> E1000_ICR_OTHER is set in EIAC. In fact, removing this code fixe= s the >>>>>>>> issue the Linux driver runs into since 4aea7a5c5e94 ("e1000e: Av= oid >>>>>>>> receiver overrun interrupt bursts") and was worked around by >>>>>>>> 745d0bd3af99 ("e1000e: Remove Other from EIAC"). >>>>>>>> >>>>>>>> Signed-off-by: Jan Kiszka >>>>>>>> --- >>>>>>>> >>>>>>>> This resolves the issue I reported on February 18 ("e1000e: MSI-= X >>>>>>>> problem with recent Linux drivers"). >>>>>>>> >>>>>>>> hw/net/e1000e_core.c | 4 ---- >>>>>>>> 1 file changed, 4 deletions(-) >>>>>>>> >>>>>>>> diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c >>>>>>>> index ecf9b15555..d38f025c0f 100644 >>>>>>>> --- a/hw/net/e1000e_core.c >>>>>>>> +++ b/hw/net/e1000e_core.c >>>>>>>> @@ -2022,10 +2022,6 @@ e1000e_msix_notify_one(E1000ECore *core, = uint32_t cause, uint32_t int_cfg) >>>>>>>> >>>>>>>> effective_eiac =3D core->mac[EIAC] & cause; >>>>>>>> >>>>>>>> - if (effective_eiac =3D=3D E1000_ICR_OTHER) { >>>>>>>> - effective_eiac |=3D E1000_ICR_OTHER_CAUSES; >>>>>>>> - } >>>>>>>> - >>>>>>>> core->mac[ICR] &=3D ~effective_eiac; >>>>>>>> >>>>>>>> if (!(core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) { >>>>>>>> >>>>>>> >>>>>>> Ping for this - as well as https://patchwork.ozlabs.org/patch/895= 476. >>>>>>> >>>>>>> Given that q35 uses e1000e by default and many Linux kernel versi= ons no >>>>>>> longer work, this should likely go into upcoming and stable versi= ons >>>>>> >>>>>> I'd rather not put it into 2.12 at this point in the release >>>>>> cycle unless it's a regression from 2.11, I think. >>>>> >>>>> Second ping - nothing hit the repo so far, nor did I receive feedba= ck. >>>>> >>>> >>>> And another ping. For both. >>>> >>>> These days I had to help someone with a broken QEMU setup that faile= d >>>> installing from network. It turned out that "modprobe e1000e IntMode= =3D0" >>>> was needed to workaround the issues my patches address. >>> >>> What about the IMS register? It is set just after. >>> >>> Looking at b38636b8372, can you test this patch? >>> >>> -- >8 -- >>> diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c >>> index c93c4661ed..a484b68a5a 100644 >>> --- a/hw/net/e1000e_core.c >>> +++ b/hw/net/e1000e_core.c >>> @@ -2022,13 +2022,13 @@ e1000e_msix_notify_one(E1000ECore *core, >>> uint32_t cause, uint32_t int_cfg) >>> >>> effective_eiac =3D core->mac[EIAC] & cause; >>> >>> - if (effective_eiac =3D=3D E1000_ICR_OTHER) { >>> - effective_eiac |=3D E1000_ICR_OTHER_CAUSES; >>> - } >>> - >>> core->mac[ICR] &=3D ~effective_eiac; >>> >>> if (!(core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) { >>> + if (effective_eiac =3D=3D E1000_ICR_OTHER) { >>> + effective_eiac |=3D E1000_ICR_OTHER_CAUSES; >>> + } >>> + >>> core->mac[IMS] &=3D ~effective_eiac; >>> } >>> } >>> >> >> Before testing this: What would be the reasoning for this change? >=20 > Not breaking the purpose of b38636b8372 :) I disagree on this expansion of bit 31 ("other causes"). I see no indication in the spec that setting this bit for autoclear has more impact than on the very same bit itself. Therefore I'm asking for a reasoning - based on the spec. Jan --=20 Siemens AG, Corporate Technology, CT RDA IOT SES-DE Corporate Competence Center Embedded Linux